Classification of Cervical Cancer Using Hybrid Multi-layered Perceptron Network Trained by Genetic Algorithm

• Published in: Procedia computer science 2019, Vol.163, p.494-501
• Main Authors: Zorkafli, Mohd Fakharuddin, Osman, Muhammad Khusairi, Isa, Iza Sazanita, Ahmad, Fadzil, Sulaiman, Siti Noraini
• Format: Article
• Language: English
• Subjects: Adaptive Fuzzy K-means clustering; cervical cancer diagnosis system; Genetic Algorithm; HMLP; HRBF; Moving K-means; Pap test
• ISSN: 1877-0509; 1877-0509
• DOI: 10.1016/j.procs.2019.12.132

Cervical cancer is well known as the third killer for women in Malaysia. The precancerous stage for detection can be determine by screening test, known Pap smear test for avoiding occurrence of cervical cancer. The problem face in the test is the human error in reading the data analysis and also lack number of pathologists to interpret the data analysis. Creating a computer-aided diagnosis system is one of the solutions that can interpret the data. Most of the research available used artificial neural network for diagnosis system to classify the cervical cancer cells data into normal and abnormal. This research creates a neural network (NN) using Hybrid Multi-layered Perceptron (HMLP) trained by Genetic Algorithm (GA) to diagnose the data. The data is extracted from cervical cells and divided into four features as the input, which are size of nucleus, size of cytoplasm, grey level of nucleus, and grey level of cytoplasm. The data is interpreted into three categories; are normal, Low-grade Squamous Intraepithelial Lesion (LSIL) and High-grade Squamous Intraepithelial Lesion (HSIL). These categories will be inserted in to the algorithm to calculate and determined the neural network performance. The data is randomly separated into dataset using 5-fold cross validation technique. The performance is compared with Hybrid Radial Basis Function (HRBF) trained with Adaptive Fuzzy K-means and Moving K-means Clustering Algorithm. This researched shows HMLP trained with GA create a better performance of the network in the accuracy, sensitivity, and specificity to be implemented in the cervical cancer for test the performance improvement.